bench: align bench_hash to backend 4-way scan path

bench/bench_hash.c

@@ -0,0 +1,59 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <time.h>
+
+#include "../sha256/sha256_backend.h"
+
+static double now_seconds(void) {
+    struct timespec ts;
+    clock_gettime(CLOCK_MONOTONIC, &ts);
+    return (double)ts.tv_sec + (double)ts.tv_nsec / 1e9;
+}
+
+int main(int argc, char **argv) {
+    uint8_t header_76[76];
+    sha256d80_midstate_t mid;
+    sha256_state_t states[4];
+    uint32_t target_words[8];
+    uint32_t nonce = 0;
+    uint64_t hashes = 0;
+    volatile uint32_t sink = 0;
+    double seconds = 10.0;
+    double t0;
+    int i;
+
+    if (argc > 1) {
+        seconds = strtod(argv[1], NULL);
+        if (seconds <= 0.0) {
+            seconds = 10.0;
+        }
+    }
+
+    for (i = 0; i < 76; i++) {
+        header_76[i] = (uint8_t)(i * 13 + 7);
+    }
+    for (i = 0; i < 8; i++) {
+        target_words[i] = 0xFFFFFFFFU;
+    }
+
+    sha256d80_midstate_init(&mid, header_76);
+
+    t0 = now_seconds();
+    while ((now_seconds() - t0) < seconds) {
+        uint32_t mask = sha256d80_scan_4way(&mid, nonce, target_words, states);
+
+        sink ^= states[0].h[0] ^ states[1].h[0] ^ states[2].h[0] ^ states[3].h[0] ^ mask;
+        nonce += 4U;
+        hashes += 4U;
+    }
+
+    {
+        double elapsed = now_seconds() - t0;
+        double khs = (double)hashes / elapsed / 1000.0;
+        printf("bench_hash: %.2f kH/s | hashes=%llu | sink=%08x\n", khs, (unsigned long long)hashes, sink);
+    }
+
+    return 0;
+}